Ara Darzi: AI could speed up the fight against antibiotic resistance, but the market stands in the way

British surgeon Ara Darzi says artificial intelligence can noticeably speed up the diagnosis of antibiotic-resistant infections and the search for new drugs. But that is not enough: unless the economics of the industry change, many developments will never reach actual clinics.

Why the crisis is accelerating

Antibiotic resistance is no longer a narrow concern for infectious disease specialists. According to Darzi, it causes more than a million deaths worldwide every year and plays a significant contributing role in almost five million more cases. These infections are harder and costlier to treat, patients stay in hospital longer, and hospitals bear extra costs.

Two things are increasing the pressure at once: excessive and improper use of antibiotics, and a weak flow of new drugs to market. Darzi called 2026 the first real turning point in this story, because the scale of the threat is now hard to ignore. A 2024 forecast published in The Lancet expects up to 40 million deaths from drug-resistant infections by 2050.

Meanwhile, in everyday practice doctors still often act almost blind. Classic diagnostics usually take two to three days: bacteria have to be grown from a sample and tested to see which drugs work. For conditions such as sepsis that is far too long: every hour of delayed treatment raises the risk of death by 4–9 percent.

Where AI is already useful

This is where AI can have the fastest effect. Darzi says AI diagnostics already show accuracy above 99 percent without additional laboratory infrastructure. That matters not only for large clinics but also for remote regions, where access to rapid tests is limited.

According to WHO estimates, in 2023 the highest levels of antibiotic resistance were recorded in Southeast Asia and the Eastern Mediterranean, where one in three reported infections was resistant. In Africa, it was one in five.

• Telling resistant infections from ordinary ones faster
• Suggesting a more accurate choice of therapy to the doctor in the first hours
• Finding new mechanisms of bacterial resistance
• Speeding up the search for and design of molecules that do not exist in nature

Another example is the work of Britain's NHS with Google DeepMind. In one demonstration, the system identified, in 48 hours, previously unknown resistance mechanisms that researchers at Imperial College London had spent about ten years working to understand. Paired with an automated laboratory, such systems can already run hundreds of parallel experiments around the clock. Deep learning models can scan billions of molecular structures within days, while generative models can propose new compounds that simply did not exist before.

Why the market gets in the way

According to Darzi, the main brake is not in the lab but in pharma's business model. New antibiotics cannot be sold as a mass-market drug: the more they are used, the faster bacteria learn to get around them.

This creates a paradox. From a medical standpoint, the best new drugs should be conserved and prescribed rarely, while commercially, companies earn money precisely from sales volume. As a result, large pharmaceutical companies have wound down their antibiotic programs in recent years, even as scientific progress in the field continued.

Breaking this logic requires different payment schemes. Darzi recalled that in 2024 the UK launched a pilot Netflix-style model: the government pays a pharmaceutical company a fixed annual sum for access to a new antibiotic, rather than for the number of doses prescribed. Sweden is also testing a model partly decoupled from volume. The idea is simple: reward not mass sales but the very fact that a vital tool exists in the healthcare system.

"We already have the tools.

The question is whether we have the character to take seriously what we are seeing."

What this means

The antibiotic resistance story shows something important: for medical AI, the bottleneck is not only model quality but also the healthcare system's ability to build these tools into practice. If diagnostics really shrink from several days to hours, and the search for new antibiotics speeds up several-fold, the winners will be not only doctors but patients.

But without new procurement rules and incentives, AI risks remaining an impressive conference demo rather than a working tool in hospitals.